Video Multi-target Tracking Using Particle Filter Technology

Video target tracking mainly achieve of the target object’s state of motion estimationin video, by estimating each frame in the image sequence of the target object to completethe position and trajectory of the target estimation in the video. In statistics, as a dynamicestimation problem, video target tracking problem is interpreted as a problem solving theposterior density function.This article mainly research the problem of video target tracking based on particlefiltering algorithm. Particle filtering is a kind of state estimation algorithm based onBayesian framework, mainly composed by the Bayesian estimation, Monte Carlointegration, sequential importance sampling and resampling technique. It is often used todeal with some nonlinear and non-gaussian problems. In video tracking problems, theparticle filter is to estimate the object state in the image through some observations imagefeatures. Through these observations to estimate the posterior probability density of thetarget. This article introduced the particle filter algorithm based on RGB color values andbased on the color characteristics of the target model respectively. And compares thetracking performance of both, Relatively speaking, Particle filter algorithm based on RGBcolor values has the advantages of simple structure and easy implement, Particle filterbased on the color characteristics of the target model has the advantages of stronganti-jamming capability, good algorithm stability, but the calculated amount is very large.In order to reduce the calculated amount of the particle filter, we propose a particlenumber of adaptive algorithms, mainly through the process of image pre-processingmethods. To slove the "sample impoverishment" in traditional resampling technique. Thispaper presents a resampling technique based on scatter search algorithm (Scatter Search,SS), scatter search algorithm generate new solutions by merger process and improveprocess that according to the reference sets, and to achieve the optimal solutions throughthe selection process. So scatter search resampling algorithm ensures both the efficientreplication of the high weight particles in the particle update process, but also ensures thediversity of the sample.Finally, this paper has carried the experiment on the particle filter algorithmbased on RGB color values and color features target model by tracking the movement of the human face, and analyzed the performance effect of system resampling and scatter search in the experiment. and it discussed the multiple target trackingproblem in video at the end of the article.